@copilotkit/aimock 1.45.1 → 1.47.0

package/dist/ollama.cjs

@@ -1,20 +1,27 @@
+const require_errors = require('./response/errors.cjs');
+const require_policy = require('./response/policy.cjs');
 const require_helpers = require('./helpers.cjs');
 const require_sse_writer = require('./sse-writer.cjs');
 const require_chaos = require('./chaos.cjs');
 const require_misbehavior = require('./misbehavior.cjs');
+const require_embedding_request = require('./embedding-request.cjs');
 const require_router = require('./router.cjs');
 const require_recorder = require('./recorder.cjs');
-const require_interruption = require('./interruption.cjs');
+const require_timing = require('./response/timing.cjs');
+const require_deliver = require('./response/deliver.cjs');
+const require_proxy = require('./response/proxy.cjs');
 const require_ndjson_writer = require('./ndjson-writer.cjs');
+const require_raw = require('./response/raw.cjs');
+const require_ollama = require('./response/usage/codecs/ollama.cjs');
 
 //#region src/ollama.ts
 const DURATION_FIELDS = {
 	done_reason: "stop",
 	total_duration: 0,
 	load_duration: 0,
-	prompt_eval_count: 0,
+	...require_ollama.render({ kind: "ordinary-zero" }, "prompt"),
 	prompt_eval_duration: 0,
-	eval_count: 0,
+	...require_ollama.render({ kind: "ordinary-zero" }, "completion"),
 	eval_duration: 0
 };
 function ollamaToCompletionRequest(req) {
@@ -326,8 +333,10 @@ function prepareOllamaMisbehavior(plan, request, stream, chunkSize, effectiveStr
 			for (const block of outcome.ordered) if (block.type === "text") priorContent += block.text;
 			else if (index++ === plan.target?.index) break;
 		}
+		const chunks = stream ? [...buildOllamaChatTextChunks(priorContent, model, chunkSize).slice(0, -1), { error }] : [{ error }];
 		return {
-			chunks: stream ? [...buildOllamaChatTextChunks(priorContent, model, chunkSize).slice(0, -1), { error }] : [{ error }],
+			build: () => chunks,
+			reasoning: void 0,
 			status: stream ? 200 : 500,
 			summary: {
 				...plan.summary,
@@ -343,61 +352,64 @@ function prepareOllamaMisbehavior(plan, request, stream, chunkSize, effectiveStr
 			arguments: parsed.text
 		};
 	});
-	let chunks;
-	let reasoning = "";
-	if (plan.summary.fault === "reasoning-only") {
-		reasoning = plan.reasoning ?? "";
-		const base = {
-			model,
-			created_at: (/* @__PURE__ */ new Date()).toISOString()
-		};
-		if (stream) {
-			chunks = [];
-			for (let i = 0; i < reasoning.length; i += chunkSize) chunks.push({
+	const reasoning = plan.summary.fault === "reasoning-only" ? plan.reasoning ?? "" : require_helpers.resolveReasoningForModel("reasoning" in response ? response.reasoning : void 0, model, effectiveStrict, logger) ?? "";
+	const completion = content + reasoning + servedToolCalls.map((call) => call.name + call.arguments).join("");
+	const build = () => {
+		let chunks;
+		if (plan.summary.fault === "reasoning-only") {
+			const base = {
+				model,
+				created_at: (/* @__PURE__ */ new Date()).toISOString()
+			};
+			if (stream) {
+				chunks = [];
+				for (let i = 0; i < reasoning.length; i += chunkSize) chunks.push({
+					...base,
+					message: {
+						role: "assistant",
+						content: "",
+						thinking: reasoning.slice(i, i + chunkSize)
+					},
+					done: false
+				});
+				chunks.push({
+					...base,
+					message: {
+						role: "assistant",
+						content: ""
+					},
+					done: true,
+					...DURATION_FIELDS,
+					done_reason: "length"
+				});
+			} else chunks = [{
 				...base,
 				message: {
 					role: "assistant",
 					content: "",
-					thinking: reasoning.slice(i, i + chunkSize)
-				},
-				done: false
-			});
-			chunks.push({
-				...base,
-				message: {
-					role: "assistant",
-					content: ""
+					thinking: reasoning
 				},
 				done: true,
 				...DURATION_FIELDS,
 				done_reason: "length"
-			});
-		} else chunks = [{
-			...base,
-			message: {
-				role: "assistant",
-				content: "",
-				thinking: reasoning
-			},
-			done: true,
-			...DURATION_FIELDS,
-			done_reason: "length"
-		}];
-	} else {
-		reasoning = require_helpers.resolveReasoningForModel("reasoning" in response ? response.reasoning : void 0, model, effectiveStrict, logger) ?? "";
-		if (combined) chunks = stream ? buildOllamaChatContentWithToolCallsChunks(content, calls, model, chunkSize, reasoning, response.blocks) : [buildOllamaChatContentWithToolCallsResponse(content, calls, model, reasoning, response.blocks)];
+			}];
+		} else if (combined) chunks = stream ? buildOllamaChatContentWithToolCallsChunks(content, calls, model, chunkSize, reasoning, response.blocks) : [buildOllamaChatContentWithToolCallsResponse(content, calls, model, reasoning, response.blocks)];
 		else if (require_helpers.isToolCallResponse(response)) chunks = stream ? buildOllamaChatToolCallChunks(calls, model, chunkSize, reasoning) : [buildOllamaChatToolCallResponse(calls, model, reasoning)];
 		else chunks = stream ? buildOllamaChatTextChunks(content, model, chunkSize, reasoning) : [buildOllamaChatTextResponse(content, model, reasoning)];
-	}
-	const completion = content + reasoning + servedToolCalls.map((call) => call.name + call.arguments).join("");
-	const terminal = chunks.length - 1;
-	chunks[terminal] = {
-		...chunks[terminal],
-		prompt_eval_count: require_helpers.estimatePromptTokens(request.messages),
-		eval_count: require_helpers.estimateTokens(completion)
+		const terminal = chunks.length - 1;
+		chunks[terminal] = {
+			...chunks[terminal],
+			...require_ollama.render({
+				kind: "applied-chat-estimate",
+				messages: request.messages,
+				completion
+			})
+		};
+		return chunks;
 	};
 	return {
-		chunks,
+		build,
+		reasoning,
 		status: 200,
 		summary: {
 			...plan.summary,
@@ -467,10 +479,10 @@ async function handleOllama(req, res, raw, fixtures, journal, defaults, setCorsH
 				fixture: null
 			}
 		});
-		require_sse_writer.writeErrorResponse(res, 400, JSON.stringify({ error: {
+		require_sse_writer.writeErrorResponse(res, 400, require_errors.renderError("ollama", "validation", {
 			message: `Malformed JSON body: ${detail}`,
 			type: "invalid_request_error"
-		} }));
+		}).body);
 		return;
 	}
 	if (!require_helpers.isJsonObject(ollamaReq)) {
@@ -484,10 +496,10 @@ async function handleOllama(req, res, raw, fixtures, journal, defaults, setCorsH
 				fixture: null
 			}
 		});
-		require_sse_writer.writeErrorResponse(res, 400, JSON.stringify({ error: {
+		require_sse_writer.writeErrorResponse(res, 400, require_errors.renderError("ollama", "validation", {
 			message: "Request body must be a JSON object",
 			type: "invalid_request_error"
-		} }));
+		}).body);
 		return;
 	}
 	let chatShapeError = require_helpers.validateChatMessages(ollamaReq.messages) ?? require_helpers.validateToolsField(ollamaReq.tools);
@@ -506,17 +518,20 @@ async function handleOllama(req, res, raw, fixtures, journal, defaults, setCorsH
 				fixture: null
 			}
 		});
-		require_sse_writer.writeErrorResponse(res, 400, JSON.stringify({ error: {
+		require_sse_writer.writeErrorResponse(res, 400, require_errors.renderError("ollama", "validation", {
 			message: `Invalid request: ${chatShapeError}`,
 			type: "invalid_request_error"
-		} }));
+		}).body);
 		return;
 	}
 	const completionReq = ollamaToCompletionRequest(ollamaReq);
 	completionReq._endpointType = "chat";
 	completionReq._context = require_helpers.getContext(req);
 	const testId = require_helpers.getTestId(req);
-	const { fixture, skippedBySequenceOrTurn } = require_router.matchFixtureDiagnostic(fixtures, completionReq, journal.getFixtureMatchCountsForTest(testId), defaults.requestTransform, require_router.recordMatchOptions(!!defaults.record, defaults.logger));
+	const { fixture, skippedBySequenceOrTurn } = require_router.matchFixtureDiagnostic(fixtures, completionReq, journal.getFixtureMatchCountsForTest(testId), defaults.requestTransform, {
+		...require_router.recordMatchOptions(!!defaults.record, defaults.logger),
+		wire: "ollama"
+	});
 	if (fixture) {
 		journal.incrementFixtureMatchCount(fixture, fixtures, testId);
 		logger.debug(`Fixture matched: ${JSON.stringify(fixture.match).slice(0, 120)}`);
@@ -528,7 +543,7 @@ async function handleOllama(req, res, raw, fixtures, journal, defaults, setCorsH
 		body: completionReq
 	}, fixture ? "fixture" : "proxy", defaults.registry, defaults.logger)) return;
 	if (!fixture) {
-		if (require_helpers.resolveStrictMode(defaults.strict, req.headers)) {
+		if (require_policy.resolveRequestDecisions(defaults.strict, req.headers).fixtureMiss) {
 			const strictStatus = 503;
 			const strictMessage = require_helpers.strictNoMatchMessage(skippedBySequenceOrTurn);
 			logger.error(require_helpers.strictNoMatchLogLine(req.method ?? "POST", urlPath, skippedBySequenceOrTurn));
@@ -540,17 +555,17 @@ async function handleOllama(req, res, raw, fixtures, journal, defaults, setCorsH
 				response: {
 					status: strictStatus,
 					fixture: null,
-					...require_helpers.strictOverrideField(defaults.strict, req.headers)
+					...require_policy.strictOverrideField(defaults.strict, req.headers)
 				}
 			});
-			require_sse_writer.writeErrorResponse(res, strictStatus, JSON.stringify({ error: {
+			require_sse_writer.writeErrorResponse(res, strictStatus, require_errors.renderError("ollama", "strict-miss", {
 				message: strictMessage,
 				type: "invalid_request_error"
-			} }));
+			}).body);
 			return;
 		}
 		if (defaults.record) {
-			const outcome = await require_recorder.proxyAndRecord(req, res, completionReq, "ollama", urlPath, fixtures, defaults, raw);
+			const { outcome, clientHungUp } = await require_proxy.withHangUpDetection(req, res, () => require_recorder.proxyAndRecord(req, res, completionReq, "ollama", urlPath, fixtures, defaults, raw));
 			if (outcome === "handled_by_hook") return;
 			if (outcome !== "not_configured") {
 				journal.add({
@@ -561,7 +576,8 @@ async function handleOllama(req, res, raw, fixtures, journal, defaults, setCorsH
 					response: {
 						status: res.statusCode ?? 200,
 						fixture: null,
-						source: "proxy"
+						source: "proxy",
+						...require_proxy.proxyInterruption(res, clientHungUp)
 					}
 				});
 				return;
@@ -575,13 +591,13 @@ async function handleOllama(req, res, raw, fixtures, journal, defaults, setCorsH
 			response: {
 				status: 404,
 				fixture: null,
-				...require_helpers.strictOverrideField(defaults.strict, req.headers)
+				...require_policy.strictOverrideField(defaults.strict, req.headers)
 			}
 		});
-		require_sse_writer.writeErrorResponse(res, 404, JSON.stringify({ error: {
+		require_sse_writer.writeErrorResponse(res, 404, require_errors.renderError("ollama", "fixture-miss", {
 			message: "No fixture matched",
 			type: "invalid_request_error"
-		} }));
+		}).body);
 		return;
 	}
 	const response = await require_helpers.resolveResponse(fixture, completionReq);
@@ -613,14 +629,50 @@ async function handleOllama(req, res, raw, fixtures, journal, defaults, setCorsH
 		});
 		const message = `${misbehavior.code}: ${misbehavior.message}`;
 		if (!misbehavior.summary?.evaluations.length) defaults.logger.error(message);
-		require_sse_writer.writeErrorResponse(res, misbehavior.status, JSON.stringify({ error: message }));
+		require_sse_writer.writeErrorResponse(res, misbehavior.status, require_errors.renderError("ollama", "misbehavior", {
+			message: misbehavior.message,
+			code: misbehavior.code
+		}).body);
+		return;
+	}
+	if (require_helpers.isRawResponse(response)) {
+		await require_raw.serveRawResponse({
+			wire: "ollama",
+			stream: ollamaReq.stream !== false,
+			req,
+			res,
+			fixture,
+			response,
+			body: completionReq,
+			journal,
+			defaults,
+			testId,
+			misbehaviorSummary: misbehavior.summary,
+			errorWire: "ollama"
+		});
 		return;
 	}
 	const latency = fixture.latency ?? defaults.latency;
 	const chunkSize = Math.max(1, fixture.chunkSize ?? defaults.chunkSize);
+	const replaySpeed = fixture.replaySpeed ?? defaults.replaySpeed;
+	const streamTiming = {
+		latency,
+		streamingProfile: fixture.streamingProfile,
+		recordedTimings: fixture.recordedTimings,
+		replaySpeed
+	};
+	const jsonOpts = (payloadLength, duplicateId = false) => ({
+		surface: "other",
+		payloadLength,
+		chunkSize,
+		recordedTimings: fixture.recordedTimings,
+		duplicateId,
+		replaySpeed
+	});
 	const streaming = ollamaReq.stream !== false;
 	if (misbehavior.kind === "applied") {
-		const prepared = prepareOllamaMisbehavior(misbehavior, completionReq, streaming, chunkSize, require_helpers.resolveStrictMode(defaults.strict, req.headers), logger);
+		const prepared = prepareOllamaMisbehavior(misbehavior, completionReq, streaming, chunkSize, require_policy.resolveRequestDecisions(defaults.strict, req.headers).reasoningSuppression, logger);
+		const chunks = !streaming && prepared.status === 200 && require_timing.nonStreamingTimingEnabled(fixture, "other") ? void 0 : prepared.build();
 		const journalEntry = journal.add({
 			method: req.method ?? "POST",
 			path: urlPath,
@@ -637,24 +689,16 @@ async function handleOllama(req, res, raw, fixtures, journal, defaults, setCorsH
 			defaults,
 			testId
 		});
-		if (!streaming) {
+		if (!streaming && prepared.status !== 200) {
 			res.writeHead(prepared.status, { "Content-Type": "application/json" });
-			res.end(JSON.stringify(prepared.chunks[0]));
-		} else {
-			const interruption = require_interruption.createInterruptionSignal(fixture);
-			if (!await require_ndjson_writer.writeNDJSONStream(res, prepared.chunks, {
-				latency,
-				streamingProfile: fixture.streamingProfile,
-				recordedTimings: fixture.recordedTimings,
-				replaySpeed: fixture.replaySpeed ?? defaults.replaySpeed,
-				signal: interruption?.signal,
-				onChunkSent: interruption?.tick
-			})) {
-				if (!res.writableEnded) res.destroy();
-				journalEntry.response.interrupted = true;
-				journalEntry.response.interruptReason = interruption?.reason();
-			}
-			interruption?.cleanup();
+			res.end(JSON.stringify((chunks ?? prepared.build())[0]));
+		} else if (!streaming) await require_deliver.deliverJson(res, journalEntry, fixture, () => (chunks ?? prepared.build())[0], {
+			...jsonOpts(require_timing.misbehaviorPayloadLength(misbehavior, misbehavior.reasoning === void 0 ? prepared.reasoning : void 0), !!misbehavior.duplicateId),
+			status: prepared.status
+		});
+		else {
+			const frames = chunks ?? prepared.build();
+			await require_deliver.deliverStream(res, journalEntry, fixture, require_ndjson_writer.ndjsonCodec(), frames, streamTiming);
 		}
 		return;
 	}
@@ -675,7 +719,7 @@ async function handleOllama(req, res, raw, fixtures, journal, defaults, setCorsH
 			defaults,
 			testId
 		});
-		require_sse_writer.writeErrorResponse(res, status, require_helpers.serializeErrorResponse(response), { retryAfter: response.retryAfter });
+		require_sse_writer.writeErrorResponse(res, status, require_errors.renderError("openai", "fixture", response.error).body, { retryAfter: response.retryAfter });
 		return;
 	}
 	if (require_helpers.isCombinedFixtureResponse(response)) {
@@ -697,28 +741,12 @@ async function handleOllama(req, res, raw, fixtures, journal, defaults, setCorsH
 			testId
 		});
 		const functionToolCalls = require_helpers.requireEmittedFunctionToolCalls(response, "Ollama");
-		const effectiveStrict = require_helpers.resolveStrictMode(defaults.strict, req.headers);
+		const effectiveStrict = require_policy.resolveRequestDecisions(defaults.strict, req.headers).reasoningSuppression;
 		const effReasoning = require_helpers.resolveReasoningForModel(response.reasoning, completionReq.model, effectiveStrict, logger);
-		if (!streaming) {
-			const body = buildOllamaChatContentWithToolCallsResponse(response.content ?? "", functionToolCalls, completionReq.model, effReasoning, response.blocks);
-			res.writeHead(200, { "Content-Type": "application/json" });
-			res.end(JSON.stringify(body));
-		} else {
+		if (!streaming) await require_deliver.deliverJson(res, journalEntry, fixture, () => buildOllamaChatContentWithToolCallsResponse(response.content ?? "", functionToolCalls, completionReq.model, effReasoning, response.blocks), jsonOpts(require_timing.servedPayloadLength(response, effReasoning)));
+		else {
 			const chunks = buildOllamaChatContentWithToolCallsChunks(response.content ?? "", functionToolCalls, completionReq.model, chunkSize, effReasoning, response.blocks);
-			const interruption = require_interruption.createInterruptionSignal(fixture);
-			if (!await require_ndjson_writer.writeNDJSONStream(res, chunks, {
-				latency,
-				streamingProfile: fixture.streamingProfile,
-				recordedTimings: fixture.recordedTimings,
-				replaySpeed: fixture.replaySpeed ?? defaults.replaySpeed,
-				signal: interruption?.signal,
-				onChunkSent: interruption?.tick
-			})) {
-				if (!res.writableEnded) res.destroy();
-				journalEntry.response.interrupted = true;
-				journalEntry.response.interruptReason = interruption?.reason();
-			}
-			interruption?.cleanup();
+			await require_deliver.deliverStream(res, journalEntry, fixture, require_ndjson_writer.ndjsonCodec(), chunks, streamTiming);
 		}
 		return;
 	}
@@ -740,28 +768,12 @@ async function handleOllama(req, res, raw, fixtures, journal, defaults, setCorsH
 			defaults,
 			testId
 		});
-		const effectiveStrict = require_helpers.resolveStrictMode(defaults.strict, req.headers);
+		const effectiveStrict = require_policy.resolveRequestDecisions(defaults.strict, req.headers).reasoningSuppression;
 		const effReasoning = require_helpers.resolveReasoningForModel(response.reasoning, completionReq.model, effectiveStrict, logger);
-		if (!streaming) {
-			const body = buildOllamaChatTextResponse(response.content, completionReq.model, effReasoning);
-			res.writeHead(200, { "Content-Type": "application/json" });
-			res.end(JSON.stringify(body));
-		} else {
+		if (!streaming) await require_deliver.deliverJson(res, journalEntry, fixture, () => buildOllamaChatTextResponse(response.content, completionReq.model, effReasoning), jsonOpts(require_timing.servedPayloadLength(response, effReasoning)));
+		else {
 			const chunks = buildOllamaChatTextChunks(response.content, completionReq.model, chunkSize, effReasoning);
-			const interruption = require_interruption.createInterruptionSignal(fixture);
-			if (!await require_ndjson_writer.writeNDJSONStream(res, chunks, {
-				latency,
-				streamingProfile: fixture.streamingProfile,
-				recordedTimings: fixture.recordedTimings,
-				replaySpeed: fixture.replaySpeed ?? defaults.replaySpeed,
-				signal: interruption?.signal,
-				onChunkSent: interruption?.tick
-			})) {
-				if (!res.writableEnded) res.destroy();
-				journalEntry.response.interrupted = true;
-				journalEntry.response.interruptReason = interruption?.reason();
-			}
-			interruption?.cleanup();
+			await require_deliver.deliverStream(res, journalEntry, fixture, require_ndjson_writer.ndjsonCodec(), chunks, streamTiming);
 		}
 		return;
 	}
@@ -784,28 +796,12 @@ async function handleOllama(req, res, raw, fixtures, journal, defaults, setCorsH
 			testId
 		});
 		const functionToolCalls = require_helpers.requireEmittedFunctionToolCalls(response, "Ollama");
-		const effectiveStrict = require_helpers.resolveStrictMode(defaults.strict, req.headers);
+		const effectiveStrict = require_policy.resolveRequestDecisions(defaults.strict, req.headers).reasoningSuppression;
 		const effReasoning = require_helpers.resolveReasoningForModel(response.reasoning, completionReq.model, effectiveStrict, logger);
-		if (!streaming) {
-			const body = buildOllamaChatToolCallResponse(functionToolCalls, completionReq.model, effReasoning);
-			res.writeHead(200, { "Content-Type": "application/json" });
-			res.end(JSON.stringify(body));
-		} else {
+		if (!streaming) await require_deliver.deliverJson(res, journalEntry, fixture, () => buildOllamaChatToolCallResponse(functionToolCalls, completionReq.model, effReasoning), jsonOpts(require_timing.servedPayloadLength(response, effReasoning)));
+		else {
 			const chunks = buildOllamaChatToolCallChunks(functionToolCalls, completionReq.model, chunkSize, effReasoning);
-			const interruption = require_interruption.createInterruptionSignal(fixture);
-			if (!await require_ndjson_writer.writeNDJSONStream(res, chunks, {
-				latency,
-				streamingProfile: fixture.streamingProfile,
-				recordedTimings: fixture.recordedTimings,
-				replaySpeed: fixture.replaySpeed ?? defaults.replaySpeed,
-				signal: interruption?.signal,
-				onChunkSent: interruption?.tick
-			})) {
-				if (!res.writableEnded) res.destroy();
-				journalEntry.response.interrupted = true;
-				journalEntry.response.interruptReason = interruption?.reason();
-			}
-			interruption?.cleanup();
+			await require_deliver.deliverStream(res, journalEntry, fixture, require_ndjson_writer.ndjsonCodec(), chunks, streamTiming);
 		}
 		return;
 	}
@@ -824,10 +820,10 @@ async function handleOllama(req, res, raw, fixtures, journal, defaults, setCorsH
 		defaults,
 		testId
 	});
-	require_sse_writer.writeErrorResponse(res, 500, JSON.stringify({ error: {
+	require_sse_writer.writeErrorResponse(res, 500, require_errors.renderError("ollama", "unknown-response", {
 		message: "Fixture response did not match any known type",
 		type: "server_error"
-	} }));
+	}).body);
 }
 async function handleOllamaGenerate(req, res, raw, fixtures, journal, defaults, setCorsHeaders) {
 	setCorsHeaders(res);
@@ -847,10 +843,10 @@ async function handleOllamaGenerate(req, res, raw, fixtures, journal, defaults,
 				fixture: null
 			}
 		});
-		require_sse_writer.writeErrorResponse(res, 400, JSON.stringify({ error: {
+		require_sse_writer.writeErrorResponse(res, 400, require_errors.renderError("ollama", "validation", {
 			message: `Malformed JSON body: ${detail}`,
 			type: "invalid_request_error"
-		} }));
+		}).body);
 		return;
 	}
 	if (!require_helpers.isJsonObject(generateReq)) {
@@ -864,10 +860,10 @@ async function handleOllamaGenerate(req, res, raw, fixtures, journal, defaults,
 				fixture: null
 			}
 		});
-		require_sse_writer.writeErrorResponse(res, 400, JSON.stringify({ error: {
+		require_sse_writer.writeErrorResponse(res, 400, require_errors.renderError("ollama", "validation", {
 			message: "Request body must be a JSON object",
 			type: "invalid_request_error"
-		} }));
+		}).body);
 		return;
 	}
 	if (!generateReq.prompt || typeof generateReq.prompt !== "string") {
@@ -881,17 +877,20 @@ async function handleOllamaGenerate(req, res, raw, fixtures, journal, defaults,
 				fixture: null
 			}
 		});
-		require_sse_writer.writeErrorResponse(res, 400, JSON.stringify({ error: {
+		require_sse_writer.writeErrorResponse(res, 400, require_errors.renderError("ollama", "validation", {
 			message: "Invalid request: prompt field is required",
 			type: "invalid_request_error"
-		} }));
+		}).body);
 		return;
 	}
 	const completionReq = ollamaGenerateToCompletionRequest(generateReq);
 	completionReq._endpointType = "chat";
 	completionReq._context = require_helpers.getContext(req);
 	const testId = require_helpers.getTestId(req);
-	const { fixture, skippedBySequenceOrTurn } = require_router.matchFixtureDiagnostic(fixtures, completionReq, journal.getFixtureMatchCountsForTest(testId), defaults.requestTransform, require_router.recordMatchOptions(!!defaults.record, defaults.logger));
+	const { fixture, skippedBySequenceOrTurn } = require_router.matchFixtureDiagnostic(fixtures, completionReq, journal.getFixtureMatchCountsForTest(testId), defaults.requestTransform, {
+		...require_router.recordMatchOptions(!!defaults.record, defaults.logger),
+		wire: "ollama-generate"
+	});
 	if (fixture) {
 		journal.incrementFixtureMatchCount(fixture, fixtures, testId);
 		defaults.logger.debug(`Fixture matched: ${JSON.stringify(fixture.match).slice(0, 120)}`);
@@ -903,7 +902,7 @@ async function handleOllamaGenerate(req, res, raw, fixtures, journal, defaults,
 		body: completionReq
 	}, fixture ? "fixture" : "proxy", defaults.registry, defaults.logger)) return;
 	if (!fixture) {
-		if (require_helpers.resolveStrictMode(defaults.strict, req.headers)) {
+		if (require_policy.resolveRequestDecisions(defaults.strict, req.headers).fixtureMiss) {
 			const strictStatus = 503;
 			const strictMessage = require_helpers.strictNoMatchMessage(skippedBySequenceOrTurn);
 			defaults.logger.error(require_helpers.strictNoMatchLogLine(req.method ?? "POST", urlPath, skippedBySequenceOrTurn));
@@ -915,17 +914,17 @@ async function handleOllamaGenerate(req, res, raw, fixtures, journal, defaults,
 				response: {
 					status: strictStatus,
 					fixture: null,
-					...require_helpers.strictOverrideField(defaults.strict, req.headers)
+					...require_policy.strictOverrideField(defaults.strict, req.headers)
 				}
 			});
-			require_sse_writer.writeErrorResponse(res, strictStatus, JSON.stringify({ error: {
+			require_sse_writer.writeErrorResponse(res, strictStatus, require_errors.renderError("ollama", "strict-miss", {
 				message: strictMessage,
 				type: "invalid_request_error"
-			} }));
+			}).body);
 			return;
 		}
 		if (defaults.record) {
-			const outcome = await require_recorder.proxyAndRecord(req, res, completionReq, "ollama", urlPath, fixtures, defaults, raw);
+			const { outcome, clientHungUp } = await require_proxy.withHangUpDetection(req, res, () => require_recorder.proxyAndRecord(req, res, completionReq, "ollama", urlPath, fixtures, defaults, raw));
 			if (outcome === "handled_by_hook") return;
 			if (outcome !== "not_configured") {
 				journal.add({
@@ -936,7 +935,8 @@ async function handleOllamaGenerate(req, res, raw, fixtures, journal, defaults,
 					response: {
 						status: res.statusCode ?? 200,
 						fixture: null,
-						source: "proxy"
+						source: "proxy",
+						...require_proxy.proxyInterruption(res, clientHungUp)
 					}
 				});
 				return;
@@ -950,13 +950,13 @@ async function handleOllamaGenerate(req, res, raw, fixtures, journal, defaults,
 			response: {
 				status: 404,
 				fixture: null,
-				...require_helpers.strictOverrideField(defaults.strict, req.headers)
+				...require_policy.strictOverrideField(defaults.strict, req.headers)
 			}
 		});
-		require_sse_writer.writeErrorResponse(res, 404, JSON.stringify({ error: {
+		require_sse_writer.writeErrorResponse(res, 404, require_errors.renderError("ollama", "fixture-miss", {
 			message: "No fixture matched",
 			type: "invalid_request_error"
-		} }));
+		}).body);
 		return;
 	}
 	const response = await require_helpers.resolveResponse(fixture, completionReq);
@@ -989,11 +989,46 @@ async function handleOllamaGenerate(req, res, raw, fixtures, journal, defaults,
 		});
 		const message = `${misbehavior.code}: ${misbehavior.message}`;
 		if (!misbehavior.summary?.evaluations.length) defaults.logger.error(message);
-		require_sse_writer.writeErrorResponse(res, misbehavior.status, JSON.stringify({ error: message }));
+		require_sse_writer.writeErrorResponse(res, misbehavior.status, require_errors.renderError("ollama", "misbehavior", {
+			message: misbehavior.message,
+			code: misbehavior.code
+		}).body);
+		return;
+	}
+	if (require_helpers.isRawResponse(response)) {
+		await require_raw.serveRawResponse({
+			wire: "ollama-generate",
+			stream: generateReq.stream !== false,
+			req,
+			res,
+			fixture,
+			response,
+			body: completionReq,
+			journal,
+			defaults,
+			testId,
+			misbehaviorSummary: misbehavior.summary,
+			errorWire: "ollama"
+		});
 		return;
 	}
 	const latency = fixture.latency ?? defaults.latency;
 	const chunkSize = Math.max(1, fixture.chunkSize ?? defaults.chunkSize);
+	const replaySpeed = fixture.replaySpeed ?? defaults.replaySpeed;
+	const streamTiming = {
+		latency,
+		streamingProfile: fixture.streamingProfile,
+		recordedTimings: fixture.recordedTimings,
+		replaySpeed
+	};
+	const jsonOpts = (payloadLength, duplicateId = false) => ({
+		surface: "other",
+		payloadLength,
+		chunkSize,
+		recordedTimings: fixture.recordedTimings,
+		duplicateId,
+		replaySpeed
+	});
 	const streaming = generateReq.stream !== false;
 	if (require_helpers.isErrorResponse(response)) {
 		const status = response.status ?? 500;
@@ -1012,7 +1047,7 @@ async function handleOllamaGenerate(req, res, raw, fixtures, journal, defaults,
 			defaults,
 			testId
 		});
-		require_sse_writer.writeErrorResponse(res, status, require_helpers.serializeErrorResponse(response), { retryAfter: response.retryAfter });
+		require_sse_writer.writeErrorResponse(res, status, require_errors.renderError("openai", "fixture", response.error).body, { retryAfter: response.retryAfter });
 		return;
 	}
 	if (require_helpers.isTextResponse(response)) {
@@ -1032,28 +1067,12 @@ async function handleOllamaGenerate(req, res, raw, fixtures, journal, defaults,
 			defaults,
 			testId
 		});
-		const effectiveStrict = require_helpers.resolveStrictMode(defaults.strict, req.headers);
+		const effectiveStrict = require_policy.resolveRequestDecisions(defaults.strict, req.headers).reasoningSuppression;
 		const effReasoning = require_helpers.resolveReasoningForModel(response.reasoning, completionReq.model, effectiveStrict, defaults.logger);
-		if (!streaming) {
-			const body = buildOllamaGenerateTextResponse(response.content, completionReq.model, effReasoning);
-			res.writeHead(200, { "Content-Type": "application/json" });
-			res.end(JSON.stringify(body));
-		} else {
+		if (!streaming) await require_deliver.deliverJson(res, journalEntry, fixture, () => buildOllamaGenerateTextResponse(response.content, completionReq.model, effReasoning), jsonOpts(require_timing.servedPayloadLength({ content: response.content }, effReasoning)));
+		else {
 			const chunks = buildOllamaGenerateTextChunks(response.content, completionReq.model, chunkSize, effReasoning);
-			const interruption = require_interruption.createInterruptionSignal(fixture);
-			if (!await require_ndjson_writer.writeNDJSONStream(res, chunks, {
-				latency,
-				streamingProfile: fixture.streamingProfile,
-				recordedTimings: fixture.recordedTimings,
-				replaySpeed: fixture.replaySpeed ?? defaults.replaySpeed,
-				signal: interruption?.signal,
-				onChunkSent: interruption?.tick
-			})) {
-				if (!res.writableEnded) res.destroy();
-				journalEntry.response.interrupted = true;
-				journalEntry.response.interruptReason = interruption?.reason();
-			}
-			interruption?.cleanup();
+			await require_deliver.deliverStream(res, journalEntry, fixture, require_ndjson_writer.ndjsonCodec(), chunks, streamTiming);
 		}
 		return;
 	}
@@ -1073,10 +1092,10 @@ async function handleOllamaGenerate(req, res, raw, fixtures, journal, defaults,
 			defaults,
 			testId
 		});
-		require_sse_writer.writeErrorResponse(res, 400, JSON.stringify({ error: {
+		require_sse_writer.writeErrorResponse(res, 400, require_errors.renderError("ollama", "validation", {
 			message: "Tool call fixtures are not supported on /api/generate — use /api/chat instead",
 			type: "invalid_request_error"
-		} }));
+		}).body);
 		return;
 	}
 	require_misbehavior.recordMisbehaviorOutcome({
@@ -1094,15 +1113,17 @@ async function handleOllamaGenerate(req, res, raw, fixtures, journal, defaults,
 		defaults,
 		testId
 	});
-	require_sse_writer.writeErrorResponse(res, 500, JSON.stringify({ error: {
+	require_sse_writer.writeErrorResponse(res, 500, require_errors.renderError("ollama", "unknown-response", {
 		message: "Fixture response did not match any known type",
 		type: "server_error"
-	} }));
+	}).body);
 }
 async function handleOllamaEmbeddings(req, res, raw, fixtures, journal, defaults, setCorsHeaders) {
 	const { logger } = defaults;
 	setCorsHeaders(res);
 	const urlPath = req.url ?? "/api/embeddings";
+	const currentEndpoint = urlPath.split("?")[0] === "/api/embed";
+	const errorWire = currentEndpoint ? "ollama-embed" : "ollama";
 	let embReq;
 	try {
 		embReq = JSON.parse(raw);
@@ -1118,10 +1139,10 @@ async function handleOllamaEmbeddings(req, res, raw, fixtures, journal, defaults
 				fixture: null
 			}
 		});
-		require_sse_writer.writeErrorResponse(res, 400, JSON.stringify({ error: {
+		require_sse_writer.writeErrorResponse(res, 400, require_errors.renderError(errorWire, "validation", {
 			message: `Malformed JSON body: ${detail}`,
 			type: "invalid_request_error"
-		} }));
+		}).body);
 		return;
 	}
 	if (!require_helpers.isJsonObject(embReq)) {
@@ -1135,10 +1156,10 @@ async function handleOllamaEmbeddings(req, res, raw, fixtures, journal, defaults
 				fixture: null
 			}
 		});
-		require_sse_writer.writeErrorResponse(res, 400, JSON.stringify({ error: {
+		require_sse_writer.writeErrorResponse(res, 400, require_errors.renderError(errorWire, "validation", {
 			message: "Request body must be a JSON object",
 			type: "invalid_request_error"
-		} }));
+		}).body);
 		return;
 	}
 	const promptShapeError = embReq.prompt !== void 0 && embReq.prompt !== null && typeof embReq.prompt !== "string" ? "prompt field must be a string" : embReq.input !== void 0 && embReq.input !== null && require_helpers.normalizeEmbeddingInput(embReq.input) === null ? "input field must be a string or an array of strings" : null;
@@ -1153,14 +1174,15 @@ async function handleOllamaEmbeddings(req, res, raw, fixtures, journal, defaults
 				fixture: null
 			}
 		});
-		require_sse_writer.writeErrorResponse(res, 400, JSON.stringify({ error: {
+		require_sse_writer.writeErrorResponse(res, 400, require_errors.renderError(errorWire, "validation", {
 			message: `Invalid request: ${promptShapeError}`,
 			type: "invalid_request_error"
-		} }));
+		}).body);
 		return;
 	}
-	const inputText = embReq.prompt ?? (typeof embReq.input === "string" ? embReq.input : void 0) ?? (Array.isArray(embReq.input) ? embReq.input.join(" ") : void 0);
-	if (!embReq.model) {
+	const inputs = require_helpers.normalizeEmbeddingInput(currentEndpoint ? embReq.input ?? embReq.prompt : embReq.prompt ?? embReq.input);
+	const inputText = currentEndpoint ? inputs?.join(" ") : embReq.prompt ?? (typeof embReq.input === "string" ? embReq.input : void 0) ?? (Array.isArray(embReq.input) ? embReq.input.join(" ") : void 0);
+	if (!embReq.model || currentEndpoint && typeof embReq.model !== "string") {
 		journal.add({
 			method: req.method ?? "POST",
 			path: urlPath,
@@ -1171,10 +1193,10 @@ async function handleOllamaEmbeddings(req, res, raw, fixtures, journal, defaults
 				fixture: null
 			}
 		});
-		require_sse_writer.writeErrorResponse(res, 400, JSON.stringify({ error: {
+		require_sse_writer.writeErrorResponse(res, 400, require_errors.renderError(errorWire, "validation", {
 			message: "Invalid request: model field is required",
 			type: "invalid_request_error"
-		} }));
+		}).body);
 		return;
 	}
 	if (!inputText) {
@@ -1188,10 +1210,10 @@ async function handleOllamaEmbeddings(req, res, raw, fixtures, journal, defaults
 				fixture: null
 			}
 		});
-		require_sse_writer.writeErrorResponse(res, 400, JSON.stringify({ error: {
+		require_sse_writer.writeErrorResponse(res, 400, require_errors.renderError(errorWire, "validation", {
 			message: "Invalid request: prompt or input field is required",
 			type: "invalid_request_error"
-		} }));
+		}).body);
 		return;
 	}
 	const syntheticReq = {
@@ -1201,6 +1223,11 @@ async function handleOllamaEmbeddings(req, res, raw, fixtures, journal, defaults
 		_endpointType: "embedding",
 		_context: require_helpers.getContext(req)
 	};
+	require_embedding_request.setEmbeddingRequest(syntheticReq, {
+		operation: currentEndpoint ? "/api/embed" : "/api/embeddings",
+		input: currentEndpoint ? embReq.input ?? embReq.prompt ?? "" : embReq.prompt ?? embReq.input ?? "",
+		...embReq.dimensions !== void 0 && { dimensions: embReq.dimensions }
+	});
 	const testId = require_helpers.getTestId(req);
 	const { fixture, skippedBySequenceOrTurn } = require_router.matchFixtureDiagnostic(fixtures, syntheticReq, journal.getFixtureMatchCountsForTest(testId), defaults.requestTransform, require_router.recordMatchOptions(!!defaults.record, defaults.logger));
 	if (fixture) {
@@ -1227,7 +1254,23 @@ async function handleOllamaEmbeddings(req, res, raw, fixtures, journal, defaults
 					fixture
 				}
 			});
-			require_sse_writer.writeErrorResponse(res, status, require_helpers.serializeErrorResponse(response), { retryAfter: response.retryAfter });
+			require_sse_writer.writeErrorResponse(res, status, require_errors.renderError(currentEndpoint ? "ollama-embed" : "openai", "fixture", response.error).body, { retryAfter: response.retryAfter });
+			return;
+		}
+		if (currentEndpoint && require_helpers.isJSONResponse(response)) {
+			const status = response.status ?? 200;
+			journal.add({
+				method: req.method ?? "POST",
+				path: urlPath,
+				headers: require_helpers.flattenHeaders(req.headers),
+				body: syntheticReq,
+				response: {
+					status,
+					fixture
+				}
+			});
+			res.writeHead(status, { "Content-Type": "application/json" });
+			res.end(JSON.stringify(response.json));
 			return;
 		}
 		if (require_helpers.isEmbeddingResponse(response)) {
@@ -1241,7 +1284,10 @@ async function handleOllamaEmbeddings(req, res, raw, fixtures, journal, defaults
 					fixture
 				}
 			});
-			const body = {
+			const body = currentEndpoint ? {
+				model: embReq.model,
+				embeddings: (inputs ?? []).map(() => [...response.embedding])
+			} : {
 				model: embReq.model,
 				embedding: [...response.embedding]
 			};
@@ -1259,13 +1305,13 @@ async function handleOllamaEmbeddings(req, res, raw, fixtures, journal, defaults
 				fixture
 			}
 		});
-		require_sse_writer.writeErrorResponse(res, 500, JSON.stringify({ error: {
+		require_sse_writer.writeErrorResponse(res, 500, require_errors.renderError(errorWire, "unknown-response", {
 			message: "Fixture response did not match any known embedding type (must have embedding or error)",
 			type: "server_error"
-		} }));
+		}).body);
 		return;
 	}
-	if (require_helpers.resolveStrictMode(defaults.strict, req.headers)) {
+	if (require_policy.resolveRequestDecisions(defaults.strict, req.headers).fixtureMiss) {
 		const strictMessage = require_helpers.strictNoMatchMessage(skippedBySequenceOrTurn);
 		logger.error(require_helpers.strictNoMatchLogLine(req.method ?? "POST", urlPath, skippedBySequenceOrTurn));
 		journal.add({
@@ -1276,13 +1322,13 @@ async function handleOllamaEmbeddings(req, res, raw, fixtures, journal, defaults
 			response: {
 				status: 503,
 				fixture: null,
-				...require_helpers.strictOverrideField(defaults.strict, req.headers)
+				...require_policy.strictOverrideField(defaults.strict, req.headers)
 			}
 		});
-		require_sse_writer.writeErrorResponse(res, 503, JSON.stringify({ error: {
+		require_sse_writer.writeErrorResponse(res, 503, require_errors.renderError(errorWire, "strict-miss", {
 			message: strictMessage,
 			type: "invalid_request_error"
-		} }));
+		}).body);
 		return;
 	}
 	if (defaults.record) {
@@ -1303,8 +1349,25 @@ async function handleOllamaEmbeddings(req, res, raw, fixtures, journal, defaults
 			return;
 		}
 	}
+	const dimensions = currentEndpoint ? require_helpers.validateEmbeddingDimensions(embReq.dimensions) : 1536;
+	if (dimensions === null) {
+		journal.add({
+			method: req.method ?? "POST",
+			path: urlPath,
+			headers: require_helpers.flattenHeaders(req.headers),
+			body: syntheticReq,
+			response: {
+				status: 400,
+				fixture: null
+			}
+		});
+		require_sse_writer.writeErrorResponse(res, 400, require_errors.renderError(errorWire, "validation", {
+			message: `Invalid request: dimensions must be an integer between 1 and ${require_helpers.MAX_EMBEDDING_DIMENSIONS}`,
+			type: "invalid_request_error"
+		}).body);
+		return;
+	}
 	logger.warn(`No embedding fixture matched for "${inputText.slice(0, 80)}" — returning deterministic fallback`);
-	const embedding = require_helpers.generateDeterministicEmbedding(inputText);
 	journal.add({
 		method: req.method ?? "POST",
 		path: urlPath,
@@ -1315,9 +1378,12 @@ async function handleOllamaEmbeddings(req, res, raw, fixtures, journal, defaults
 			fixture: null
 		}
 	});
-	const body = {
+	const body = currentEndpoint ? {
+		model: embReq.model,
+		embeddings: (inputs ?? []).map((input) => require_helpers.generateDeterministicEmbedding(input, dimensions))
+	} : {
 		model: embReq.model,
-		embedding
+		embedding: require_helpers.generateDeterministicEmbedding(inputText)
 	};
 	res.writeHead(200, { "Content-Type": "application/json" });
 	res.end(JSON.stringify(body));